fix(gateway): keep /model and /queue non-interrupting and execute queued /model post-turn by HearthCore · Pull Request #6252 · NousResearch/hermes-agent

HearthCore · 2026-04-08T18:37:22Z

What does this PR do?

Fixes gateway behavior for slash commands sent during an active session:

/model and /queue no longer trigger active-turn interrupts.
queued /model now executes after the current turn (via normal command pipeline) instead of being dropped.
fixes a related NameError in the first-response-before-queued-message send path.

Related Issue

Fixes #5057

Additional investigation/comment:
#5057 (comment)

Type of Change

🐛 Bug fix (non-breaking change that fixes an issue)
✅ Tests (adding or improving test coverage)

Changes Made

gateway/platforms/base.py
- extended active-session bypass allowlist with model, queue, q
gateway/run.py
- queue /model during active run (instead of hard reject)
- in _dequeue_pending_text(), keep queued slash commands as MessageEvents by re-queueing recognized commands and returning None
- fixed undefined event reference in post-turn send path
tests/gateway/test_command_bypass_active_session.py
- added regression tests for /model and /queue bypass
- added test ensuring queued slash command events are preserved for post-turn dispatch

How to Test

Start gateway and trigger a long-running turn in Telegram.
While running, send /model and /queue <prompt>.
Verify no interrupt occurs.
Verify queued /model executes after turn completion.
Run tests:
- ./venv/bin/pytest tests/gateway/test_command_bypass_active_session.py tests/gateway/test_queue_consumption.py -q

Checklist

Code

I've read the Contributing Guide
My commit messages follow Conventional Commits
My PR contains only changes related to this fix
I've tested manually on my platform: Debian 13 / Telegram gateway
I've run pytest tests/ -q and all tests pass
I've added tests for my changes

Documentation & Housekeeping

N/A (no config keys, no docs/API schema changes)

Screenshots / Logs

Manual live test result in Telegram: /model and /queue no longer interrupt active turn; queued /model executes post-turn.

/model and /queue were not listed in the active-session guard bypass in base.py, causing them to interrupt the running agent just like any plain text message would. The root cause: base.py line ~1168 has an explicit allowlist of commands that skip the interrupt path. /model and /queue were missing from it, so they fell through to the generic interrupt handler (pending_messages + active_sessions.set()), killing the running agent before run.py ever got to handle them. Fix: add 'model', 'queue', and 'q' to the bypass list so they are dispatched inline (to run.py handlers) without triggering an interrupt. - /model: returns a post-turn info message without killing the agent - /queue: stores the prompt in pending_messages for next-turn pickup Relates to: NousResearch#5057

Instead of rejecting /model while an agent is running, store the event in adapter._pending_messages and return an acknowledgement. After the current turn finishes, base adapter logic already dequeues pending events and runs them through the normal command pipeline, so /model opens its menu naturally without interrupting active work. This matches /queue semantics and improves UX for model switching in gateway chats.

…spatch Root cause: queued slash commands were dropped by _run_agent recursion flow. Pending events were converted to plain text via _dequeue_pending_text(), then command-looking text was discarded by a safety filter to avoid leaking commands into the agent as user input. This prevented queued /model from ever reaching _handle_model_command after turn completion. Fixes: - In _dequeue_pending_text(), detect recognized slash-command events, re-queue the full MessageEvent back into adapter._pending_messages, and return None so command events stay in message-pipeline path. - Keep /model and /queue bypass behavior in base.py so they don't interrupt. - Fix NameError in first-response send path (use _progress_metadata instead of undefined event variable) that appeared after restart tests. Result: - /model during active turn is acknowledged, does not interrupt, and executes after current turn as intended. - /queue remains non-interrupting and post-turn.

… command preservation

HearthCore · 2026-04-08T19:04:54Z

Ran full suite per contributing checklist:

Command: ./venv/bin/pytest tests/ -q
Result: 109 failed, 9188 passed, 30 skipped, 1 xpassed (243s)

Notably, failures span many unrelated areas (CLI quick commands, API server, provider resolution, docker env, managed media gateways, terminal requirements), while this PR only changes:

gateway/platforms/base.py
gateway/run.py
tests/gateway/test_command_bypass_active_session.py

PR-specific regression tests are green:

./venv/bin/pytest tests/gateway/test_command_bypass_active_session.py tests/gateway/test_queue_consumption.py -q
26 passed

Also manually validated in Telegram DM after gateway restart:

/queue and /model no longer interrupt active turn
queued /model executes post-turn

Any recognized slash command now bypasses the Level-1 active-session guard instead of queueing + interrupting. A mid-run /model (or /reasoning, /voice, /insights, /title, /resume, /retry, /undo, /compress, /usage, /provider, /reload-mcp, /sethome, /reset) used to interrupt the agent AND get silently discarded by the slash-command safety net — zero-char response, dropped tool calls. Root cause: - Discord registers 41 native slash commands via tree.command(). - Only 14 were in ACTIVE_SESSION_BYPASS_COMMANDS. - The other ~15 user-facing ones fell through base.py:handle_message to the busy-session handler, which calls running_agent.interrupt() AND queues the text. - After the aborted run, gateway/run.py:9912 correctly identifies the queued text as a slash command and discards it — but the damage (interrupt + zero-char response) already happened. Fix: - should_bypass_active_session() now returns True for any resolvable slash command. ACTIVE_SESSION_BYPASS_COMMANDS stays as the subset with dedicated Level-2 handlers (documentation + tests). - gateway/run.py adds a catch-all after the dedicated handlers that returns a user-visible "agent busy — wait or /stop first" response for any other resolvable command. - Unknown text / file-path-like messages are unchanged — they still queue. Also: - gateway/platforms/discord.py logs the invoker identity on every slash command (user id + name + channel + guild) so future ghost-command reports can be triaged without guessing. Tests: - 15 new parametrized cases in test_command_bypass_active_session.py cover every previously-broken Discord slash command. - Existing tests for /stop, /new, /approve, /deny, /help, /status, /agents, /background, /steer, /update, /queue still pass. - test_steer.py's ACTIVE_SESSION_BYPASS_COMMANDS check still passes. Fixes #5057. Related: #6252, #10370, #4665.

Main upstream added a broader catch-all in _process_text_command() that rejects any recognized slash command reaching the running-agent guard with a "busy — wait or /stop first" message (fixes NousResearch#5057, NousResearch#6252, NousResearch#10370). Keep the PR's specific handlers for /yolo, /verbose, /fast, /reasoning BEFORE the catch-all — these four are state toggles that should actually run (not be rejected) mid-agent. Resolution merges both blocks: 1. Toggle handlers (PR) — run yolo/verbose/fast/reasoning immediately 2. Dedicated info handlers (main) — help/commands/profile/update 3. Catch-all (main) — graceful "busy" reject for anything else Dropped /reasoning from the catch-all's example list since it now has an explicit handler above. Verified: 151 related tests pass.

teknium1 · 2026-04-27T13:27:48Z

Thanks for the thorough investigation and fix, @HearthCore! This was a real bug and your diagnosis in #5057 was accurate.

This automated hermes-sweeper review found that the core issue has since been resolved on main by a broader fix:

Commit 632a807a (PR fix(gateway): slash commands never interrupt a running agent #12334, shipped in v2026.4.23) explicitly cites Message handling behavior when multiple messages arrive during active agent execution #5057 and fix(gateway): keep /model and /queue non-interrupting and execute queued /model post-turn #6252 as the root cause.
should_bypass_active_session() in hermes_cli/commands.py:309 was widened to return True for any resolvable slash command — not just the narrow allowlist — so /model, /queue, and every other registered command now bypass the Level-1 guard automatically.
A new catch-all in gateway/run.py returns a graceful "Agent is running — /cmd can't run mid-turn. Wait or /stop first" message for any recognized slash command that reaches the running-agent guard, eliminating the interrupt+discard path.

The narrower allowlist approach in this PR is superseded by that broader fix. Closing as implemented on main.

…earch#12334) Any recognized slash command now bypasses the Level-1 active-session guard instead of queueing + interrupting. A mid-run /model (or /reasoning, /voice, /insights, /title, /resume, /retry, /undo, /compress, /usage, /provider, /reload-mcp, /sethome, /reset) used to interrupt the agent AND get silently discarded by the slash-command safety net — zero-char response, dropped tool calls. Root cause: - Discord registers 41 native slash commands via tree.command(). - Only 14 were in ACTIVE_SESSION_BYPASS_COMMANDS. - The other ~15 user-facing ones fell through base.py:handle_message to the busy-session handler, which calls running_agent.interrupt() AND queues the text. - After the aborted run, gateway/run.py:9912 correctly identifies the queued text as a slash command and discards it — but the damage (interrupt + zero-char response) already happened. Fix: - should_bypass_active_session() now returns True for any resolvable slash command. ACTIVE_SESSION_BYPASS_COMMANDS stays as the subset with dedicated Level-2 handlers (documentation + tests). - gateway/run.py adds a catch-all after the dedicated handlers that returns a user-visible "agent busy — wait or /stop first" response for any other resolvable command. - Unknown text / file-path-like messages are unchanged — they still queue. Also: - gateway/platforms/discord.py logs the invoker identity on every slash command (user id + name + channel + guild) so future ghost-command reports can be triaged without guessing. Tests: - 15 new parametrized cases in test_command_bypass_active_session.py cover every previously-broken Discord slash command. - Existing tests for /stop, /new, /approve, /deny, /help, /status, /agents, /background, /steer, /update, /queue still pass. - test_steer.py's ACTIVE_SESSION_BYPASS_COMMANDS check still passes. Fixes NousResearch#5057. Related: NousResearch#6252, NousResearch#10370, NousResearch#4665.

HearthCore added 4 commits April 8, 2026 20:36

test(gateway): cover /model and /queue active-session bypass + queued…

9f1e239

… command preservation

teknium1 mentioned this pull request Apr 19, 2026

fix(gateway): slash commands never interrupt a running agent #12334

Merged

ypwcharles mentioned this pull request Apr 22, 2026

fix(telegram): preserve busy-session queue payloads #13913

Open

teknium1 closed this Apr 27, 2026

alt-glitch added comp/gateway Gateway runner, session dispatch, delivery P2 Medium — degraded but workaround exists type/bug Something isn't working labels Apr 30, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(gateway): keep /model and /queue non-interrupting and execute queued /model post-turn#6252

fix(gateway): keep /model and /queue non-interrupting and execute queued /model post-turn#6252
HearthCore wants to merge 4 commits into
NousResearch:mainfrom
HearthCore:fix/gateway-slash-command-no-interrupt

HearthCore commented Apr 8, 2026 •

edited

Loading

Uh oh!

HearthCore commented Apr 8, 2026

Uh oh!

teknium1 commented Apr 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

HearthCore commented Apr 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Related Issue

Type of Change

Changes Made

How to Test

Checklist

Code

Documentation & Housekeeping

Screenshots / Logs

Uh oh!

HearthCore commented Apr 8, 2026

Uh oh!

teknium1 commented Apr 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

HearthCore commented Apr 8, 2026 •

edited

Loading